Comprehensive auxiliary container grabbing and releasing method and system for storage yard

Abstract

The invention discloses a comprehensive auxiliary container grabbing and placing method and system for a storage yard, and the method comprises the steps: building a three-dimensional map of the storage yard, and obtaining the three-dimensional point cloud data of a container stacking operation region in real time in the moving process of a lifting appliance, and enabling the three-dimensional point cloud data to be used for updating the three-dimensional map of the storage yard; and in the container catching and releasing process, anti-bowling resolving and pit protection resolving are conducted according to the three-dimensional map of the storage yard, the dynamic safety distance around the container and / or the lifting appliance is calculated, and when the dynamic safety distance is smaller than a threshold value, an early warning signal is sent out. According to the invention, the problems of positioning precision, environmental perception, safety protection and the like of existing storage yard operation are solved, and a solid and reliable technical foundation is laid for intelligent upgrading and unmanned operation of ports.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent yard operations, especially a comprehensive auxiliary method and system for grasping and placing containers in the yard. Background Art

[0002] The port container yard is a key core area connecting waterway and land transportation. The efficiency and safety of the daily storage, transfer, loading and unloading operations of containers directly determine the smoothness of the entire global logistics supply chain and the comprehensive operating costs and competitiveness of the port. Inside the yard, the container grasping and placing operations performed by Rail Mounted Gantry (RMG) or Rubber Tyred Gantry (RTG) are the operations with the largest quantity, highest frequency, and also the most critical. However, for a long time, this core link has been severely restricted by the existing technologies in terms of operation accuracy, efficiency and inherent safety, and there are many pain points that need to be solved urgently.

[0003] In the traditional yard operation mode, there are significant defects in the positioning technology of the crane. The position information of the trolley and the hoist mainly relies on relative positioning or low-precision technologies such as encoders, travel limit switches, or basic Differential GPS (DGPS). This positioning method is easily affected by various factors such as the cumulative error of mechanical transmission system wear, environmental temperature changes, equipment vibration, or radio signal drift, and it is difficult to provide centimeter-level high-precision absolute position information stably for a long time. In addition, the pose changes of the containers caused by yard changes (such as settlement, etc.) will also have a great impact on the above methods. This current situation of lacking a reliable and high-precision absolute positioning reference leads to a large uncertainty in the initial alignment when the spreader grabs and places containers. Operators or automated systems must spend a large amount of additional time and energy to perform repeated calibration and "fine-tuning" through visual judgment and "inching" methods. This not only greatly lengthens the operation cycle of a single container, but also accumulatively causes huge time cost consumption, directly seriously affecting the yard operation efficiency and the ship turnover rate and overall throughput capacity of the port.

[0004] In addition, the existing systems generally lack the ability to perceive the real-time three-dimensional space of the yard environment. Automated systems often rely only on the data of the preset Terminal Operating System (TOS) for operations. However, due to the complexity of the yard environment, subtle changes in the stacking of containers (such as uneven ground, wind force, slight collisions during loading and unloading, etc.), key information such as the real three-dimensional shape of the container stack, the subtle inclination angle of the container, and whether there is lateral misalignment between adjacent containers cannot be obtained in real time and accurately. When the spreader descends into the high-level or deep-level stack for operations, this lack of perception of the surrounding environment and the container state makes the system unable to predict and respond to small deviations between containers prospectively, severely restricting the precision and unmanned level of operations.

[0005] A more serious problem lies in the high level of safety hazards and the risk of equipment / cargo damage. Due to the lack of accurate spatial perception and intelligent anti-collision warning mechanisms, yard operations face high risks: on the one hand, during the lowering and placement of containers, if there is swaying of the spreader or container, positioning deviation, or lateral misalignment of adjacent containers, it is very easy for the container to be placed to collide with the adjacent stacked containers, causing damage to the containers and cargo, and even causing equipment shutdown; on the other hand, when grabbing containers from deep stacking positions, if there is a lack of real-time monitoring of the gap between the spreader and the side wall of the stack, it may cause the spreader or the container to be grabbed to rub or collide with the side edge of the stack, causing physical damage to the equipment and containers.

[0006] In conclusion, traditional port yard operation modes can no longer meet the stringent requirements of modern ports for high efficiency, high safety, and highly automated operation. Summary of the Invention

[0007] Purpose of the invention: The purpose of this invention is to provide a comprehensive auxiliary container handling method and system for container yards, establishing a comprehensive auxiliary container handling system that includes functions such as anti-bowling, deep pit protection, and container landing assistance. This fundamentally solves the bottlenecks of existing technologies in terms of positioning accuracy, environmental perception, and safety protection, laying a solid and reliable technical foundation for the intelligent upgrading and unmanned operation of ports.

[0008] Technical solution: The present invention provides a comprehensive auxiliary container handling method for stockyards, comprising the following steps:

[0009] A 3D map of the yard is established, and 3D point cloud data of the container stacking operation area is acquired in real time during the movement of the spreader to update the 3D map of the yard.

[0010] During the container handling process, anti-bowling and deep pit protection calculations are performed based on the three-dimensional map of the yard. The dynamic safety distance around the container and / or spreader is calculated, and an early warning signal is issued when the dynamic safety distance is less than the threshold.

[0011] Furthermore, a three-dimensional map of the storage yard is established using GNSS positioning;

[0012] A lidar is installed on the crane trolley to acquire three-dimensional point cloud data of the container stacking operation area. The position of the lidar in the three-dimensional map of the yard is determined according to the relative position of the lidar and the GNSS antenna. Each frame of three-dimensional point cloud data acquired by the lidar is offset to complete the fusion of the three-dimensional point cloud data and the three-dimensional map of the yard, resulting in a dynamically updated three-dimensional map of the yard.

[0013] Furthermore, each frame of 3D point cloud data acquired by the lidar is offset to complete the fusion of the 3D point cloud data with the 3D map of the storage yard, including:

[0014] 3D point cloud data offset is performed using the ICP registration mechanism:

[0015] Using GNSS positioning data as the initial value of the external parameter, the current frame's 3D point cloud data is projected onto the global coordinate system of the 3D map of the storage yard to identify geometrically overlapping areas;

[0016] The 3D point cloud data is downsampled using an adaptive voxel mesh algorithm to adjust the point cloud density to a computational equilibrium range, thus obtaining the first point cloud data.

[0017] Construct a covariance matrix based on the first point cloud data of the overlapping region of the set;

[0018] GNSS positioning data is used as a soft constraint term, which, together with geometric residual terms, is used to construct a cost function. The transformation matrix is ​​iteratively optimized to minimize the value of the cost function.

[0019] The optimized transformation matrix is ​​applied to the current frame's 3D point cloud data to complete the 3D point cloud data offset.

[0020] By fusing each frame of offset 3D point cloud data with the 3D map of the storage yard, a dynamically updated 3D map of the storage yard is obtained.

[0021] Furthermore, the cost function is:

[0022] ;

[0023] in, It is the i-th sampling point in the first point cloud data. It is the i-th sampling point in the first point cloud data after offset. The transformation matrix; It is the Mahalanobis distance weight operator. and These are the covariance matrices of the first point cloud data and the offset first point cloud data, respectively. It is a robust kernel function; As weight; The logic for determining the value is as follows:

[0024] Semantic feature classification is performed on the first point cloud data of the overlapping region of the set, including planar point set, edge point set, and key singularity set; when When identified as a key singularity ,when When identified as an edge point ,when When it is a large area planar point .

[0025] Furthermore, principal component analysis is used to obtain the eigenvalues ​​of the first point cloud data in the neighborhood, and the linear elasticity, surface flatness, and scattering are calculated.

[0026] The first point cloud data is divided into planar point sets, edge point sets, and key singularity sets based on the linear elasticity, surface flatness, and scattering. If the scattering index of the point cloud data exceeds the first threshold, it is determined to be a key singularity; otherwise, the judgment is made based on the linear elasticity index of the point cloud data. If the linear elasticity index of the point cloud data exceeds the second threshold, it is determined to be an edge point; otherwise, the judgment is made based on the surface flatness index of the point cloud data. If the surface flatness index of the point cloud data exceeds the third threshold, it is determined to be a planar point.

[0027] Furthermore, GNSS positioning data is acquired by deploying GNSS positioning units in the yard to create a three-dimensional map of the yard;

[0028] The GNSS positioning unit includes a base station module and a rover module;

[0029] The base station module includes a first GNSS antenna, a first network switching device, a base station receiver, and a base station controller, which are used to capture multi-constellation satellite signals and forward differential data sources and reference references to the rover module;

[0030] The rover module includes a second GNSS antenna, a second network switching device, an industrial control computer, a rover receiver, and a rover controller, which are used to perform RTK calculations.

[0031] The present invention discloses a yard integrated auxiliary container handling system, comprising:

[0032] The yard 3D map creation and update unit is used to create a yard 3D map and acquire 3D point cloud data of the container stacking operation area in real time during the movement of the spreader, which is used to update the yard 3D map.

[0033] The auxiliary container grabbing and placing unit is used to perform anti-bowling and deep pit protection calculations based on the three-dimensional map of the yard during the container grabbing and placing process, calculate the dynamic safety distance around the container and / or lifting gear, and issue an early warning signal when the dynamic safety distance is less than a threshold.

[0034] Furthermore, in the yard 3D map creation and update unit, a yard 3D map is created through GNSS positioning; a lidar is installed on the crane trolley to acquire 3D point cloud data of the container stacking operation area; the position of the lidar in the yard 3D map is determined according to the relative position of the lidar and the GNSS antenna; each frame of 3D point cloud data acquired by the lidar is offset to complete the fusion of the 3D point cloud data and the yard 3D map, resulting in a dynamically updated yard 3D map;

[0035] In the auxiliary grabbing and placing box unit, the ICP registration mechanism is used for 3D point cloud data offset:

[0036] Using GNSS positioning data as the initial value of the external parameter, the current frame's 3D point cloud data is projected onto the global coordinate system of the 3D map of the storage yard to identify geometrically overlapping areas;

[0037] The 3D point cloud data is downsampled using an adaptive voxel mesh algorithm to adjust the point cloud density to a computational equilibrium range, thus obtaining the first point cloud data.

[0038] Construct a covariance matrix based on the first point cloud data of the overlapping region of the set;

[0039] GNSS positioning data is used as a soft constraint term, which, together with geometric residual terms, is used to construct a cost function. The transformation matrix is ​​iteratively optimized to minimize the value of the cost function.

[0040] The optimized transformation matrix is ​​applied to the current frame's 3D point cloud data to complete the 3D point cloud data offset.

[0041] By fusing each frame of offset 3D point cloud data with the 3D map of the storage yard, a dynamically updated 3D map of the storage yard is obtained.

[0042] The computer-readable storage medium of the present invention stores a computer program, which, when executed by a processor, implements the integrated auxiliary container grabbing and placing method in the yard.

[0043] The computer program product of the present invention includes a computer program that, when executed by a processor, implements the integrated auxiliary container grabbing and placing method for the yard.

[0044] Beneficial effects: Compared with the prior art, the advantages of the present invention are as follows:

[0045] (1) This invention significantly improves the alignment accuracy and environmental robustness of yard operations by integrating GNSS global positioning and semantic weighted point cloud registration algorithm. Its core lies in the semantic feature stripping of the point cloud at three levels of "line, surface and point" and the use of dynamic weight calculation mechanism to completely solve the problems of "geometric degradation" and "horizontal sliding error" that traditional algorithms are prone to when facing large flat surfaces of containers. By assigning the highest proportion of registration weight to the sparse but critical key features of lock holes and corner fittings, the algorithm is forced to prioritize locking the core position when calculating the residual, realizing the seamless connection from global absolute coordinate guidance to local centimeter-level precision alignment. This not only greatly reduces the interference of light fluctuations and dust blockage on perception, but also fundamentally reduces the time loss of manual visual fine adjustment, providing solid and reliable algorithm support for realizing all-weather, high-efficiency automated container grabbing and placing operations.

[0046] (2) This invention integrates GNSS positioning and laser scanning auxiliary functions into a unified system platform. The architecture is simple and the deployment is flexible. It is suitable for both newly built automated terminals and for upgrading existing traditional terminals with high cost-effectiveness.

[0047] (3) This invention can accurately sense key information such as the misalignment of adjacent containers and the edges of stacked sidewalls, and enables the current yard crane to know the position of containers placed by other yard cranes in advance, so as to carry out rapid operations, save container handling time in the yard, and break through the perception blind spots of traditional systems. The anti-bowling and deep pit protection functions of this invention can predict collision risks in real time and actively intervene, effectively avoiding container side collision accidents and deep operation collisions caused by operational deviations, fundamentally reducing the risk of equipment and cargo damage, and ensuring the continuity and safety of operations.

[0048] (4) This invention provides a centimeter-level absolute position reference for crane trolleys, carriages, and spreaders for the first time through high-precision GNSS positioning, completely solving the problems of accuracy drift and uncertainty in traditional positioning methods. Combined with the precise container placement assistance function of the stacking area scanning, the absolute position of the spreader is compared with the precise coordinates of the target container, and the optimal fine-tuning command is generated in real time. This greatly reduces the time for operators to perform visual calibration and "fine-tuning" when grabbing and placing containers, effectively shortens the single container operation cycle time, and significantly improves the efficiency of stacking yard operations and port throughput capacity. Attached Figure Description

[0049] Figure 1 This is an architectural diagram of a high-precision GNSS positioning unit according to an embodiment of the present invention.

[0050] Figure 2 This is a schematic diagram of the auxiliary gripping and releasing box according to an embodiment of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0052] Example 1

[0053] The integrated auxiliary container handling method for stockyards described in this invention includes the following steps:

[0054] A 3D map of the yard is established, and 3D point cloud data of the container stacking operation area is acquired in real time during the movement of the spreader to update the 3D map of the yard.

[0055] During container handling, anti-bowling and deep-pit protection calculations are performed based on the 3D map of the container yard. The dynamic safety distance around the container and / or spreader is calculated, and a warning signal is issued when the dynamic safety distance is less than a threshold.

[0056] Specifically, in this embodiment, a three-dimensional map of the storage yard is established using a high-precision GNSS positioning unit.

[0057] like Figure 1 The diagram shows the architecture of a high-precision GNSS positioning unit, which includes a GNSS base station module and a GNSS rover module.

[0058] The base station module is used for absolute position sensing and integrates a GNSS receiving antenna, base station receiver, control module, and network switching equipment. Its core function is to acquire signals from multiple satellite constellations in real time and encapsulate these signals into standard protocol messages through forwarding logic within the main control module. These messages contain the absolute coordinates of the base station and raw observation data, aiming to provide mobile devices with a high-precision differential data source and reference benchmark.

[0059] The mobile station module consists of a receiving antenna, a mobile receiving unit, a control unit, a switch, and an industrial control computer (IPC). It is responsible for continuously acquiring signals from multiple satellite constellations and receiving differential correction messages from the base station via a low-latency network, performing high-frequency RTK calculations. The program in the IPC is responsible for converting the calculated geographic coordinates (latitude and longitude) into local planar engineering coordinates for the yard, and further incorporates a dynamic correction algorithm based on operational feedback. This algorithm utilizes the actual container grabbing position data from crane operations to compensate for and filter ground unevenness, track settlement, and theoretical yard position deviations in real time. Finally, it publishes and stores high-precision coordinate records calibrated through multiple operations in a specific data format.

[0060] Furthermore, the steps for achieving high-precision GNSS positioning are as follows:

[0061] A1: Initialization and calibration of stockyard coordinates.

[0062] For the new work area, boundary feature points are collected by performing a box-pressing operation at the four corners of the area. The system uses the latitude and longitude information of these feature points as input and automatically generates the local plane coordinate system parameters for the area through an algorithm. Initial calibration is performed to establish a precise conversion relationship between the GNSS WGS84 coordinate system and the local engineering coordinate system of the stockyard, and all positioning data is unified under the local coordinate system of the stockyard.

[0063] A2: Coordinate system transformation and static error compensation in the stockyard.

[0064] Coordinate transformation: The system converts the original geodetic coordinates (latitude, longitude, and altitude) of the base station and the rover station into local tangent plane coordinates (ENU). Based on this, a secondary transformation is performed using the initial calibration parameters to correct the coordinate axis directions so that they are parallel to the stockpile arrangement direction (Gantry / X-axis and Trolley / Y-axis), thereby establishing the stockpile coordinate system.

[0065] Installation Deviation Correction (Static): Before the system is put into operation, for each crane, the fixed deviation of its GNSS antenna center from the center of the lifting device / trolley in terms of physical structure is measured. This "installation deviation" is deducted during the coordinate transformation stage through software parameter settings.

[0066] Explanation of principle: Once the antenna installation deviation of a single unit is corrected, the coordinate data collected by different RMG trolleys are theoretically universal in the yard coordinate system, eliminating individual differences between equipment and laying the foundation for subsequent multi-vehicle data sharing.

[0067] A3: Collection of work site coordinates and adaptive filtering update.

[0068] By utilizing the communication interface with the crane PLC, a dynamic coordinate correction mechanism based on operational behavior is established to eliminate uneven ground, minor track variations, and random measurement errors.

[0069] The system monitors the spreader status in real time. When a tack lock / lock signal is detected (i.e., the spreader is engaged with the tack lock and the locking pin is locked, at which point the physical position is closest to the actual yard level), the system automatically locks and records high-frequency GNSS coordinate data for a certain period of time (e.g., 1-2 seconds before and after locking). The GNSS RTK module calculates the centimeter-level absolute three-dimensional spatial coordinates (X, Y, Z) of the crane trolley, crane carriage, and spreader, as well as the attitude information provided by the IMU, in real time at high frequencies (e.g., above 5Hz).

[0070] Data fusion: The absolute coordinates of GNSS and the relative coordinates of the crane encoder are fused and filtered to generate highly reliable real-time absolute pose data of the spreader, which serves as the basic input for all subsequent auxiliary functions.

[0071] Gantry correction logic: When a gantry performs container grabbing in a certain yard, bay, and row, the system records the gantry's X-axis coordinate in the yard coordinate system. Using a sliding window algorithm (window size can be customized, such as the last 10 times), the system first calculates the average or median value for that row; then, by combining the statistical values ​​of all rows under that bay, it calculates the global correction coordinate for that bay.

[0072] Trolley correction logic: Also based on the pressure / lock signal, record the X and Y values ​​of the trolley in the container coordinate system. Use a sliding window to filter the historical data of the same bay and row using the median / average value to eliminate random fluctuations in a single measurement and tilt errors caused by ground potholes.

[0073] Global database update: The filtered coordinate values ​​will be considered "high confidence" data and used to update the grid coordinates in the 3D yard status diagram. Since step 2) has eliminated equipment installation differences, these corrected coordinate parameters can be directly reused by other RMGs within the same yard.

[0074] A4: Distribution of box location coordinate information.

[0075] The system continuously sends out real-time dynamically corrected coordinate data of the crane's trolley and crane trolley through a predetermined communication protocol. This ensures that the external subsystems obtain not only raw GPS data, but also high-precision operational location information that incorporates the actual physical characteristics of the stockyard (such as settlement and deformation).

[0076] Furthermore, the steps of GNSS vehicle positioning are described below from the perspective of algorithm flow.

[0077] B1: The base station continuously receives satellite signals and sends its coordinate data to the rover (for differential calculation) and IPC (for resolving the base station coordinates).

[0078] B2: The mobile station continuously receives satellite signals and positioning data from the base station and performs differential calculations, then sends the differential positioning information to the IPC.

[0079] B3: The IPC will continuously receive positioning data from the base station and the rover and convert the latitude, longitude and altitude coordinates into ENU coordinates (northeast altitude coordinates) parallel to the stockyard plane. Then, it will perform corresponding rotation and translation mapping on the coordinates to obtain the stockyard coordinate system with the X-axis parallel to the stockyard direction.

[0080] B4: The algorithm will calculate based on the calibrated coordinates and give the initial values ​​of the stockpile coordinates.

[0081] B5: During subsequent operations, the latest coordinates of the current position and current row will be updated in the yard coordinate system when a ballast signal is received.

[0082] Specifically, in this embodiment, GNSS positioning and three-dimensional laser point cloud are combined for integrated auxiliary container handling in the storage yard. For example... Figure 2 The diagram shows an auxiliary container grabbing and placing system. The gray cuboids represent the trolley, the blue ones the main trolley, the red ones the two installed radars, and the yellow ones the radar detection. Using these two radars and GNSS, the entire yard can be visualized in 3D as the trolley and main trolley move. These two radars also ensure safety during operations and assist with container grabbing and placing. Specifically, this includes:

[0083] (1) Three-dimensional scanning sensing unit of the stack area.

[0084] High-performance 3D scanning equipment: Employing high-performance LiDAR or visual scanning equipment, deployed at the optimal observation point on the crane trolley to achieve real-time, active 3D scanning of the container stacking area below. This embodiment uses non-contact laser scanning technology, which is not limited by lighting conditions and can operate stably in all weather conditions such as nighttime, rain, and snow, ensuring system reliability.

[0085] Point cloud data acquisition: Real-time capture of high-density, centimeter-level precision 3D point cloud data of the target container, adjacent containers, stacked ground, and spreader working area.

[0086] (2) Stacking status identification and safety calculation unit.

[0087] Point cloud intelligent processing module: Denoises, segments and clusters the collected point cloud data, intelligently identifies and extracts key operational information, including: the actual three-dimensional shape of the container on the target stacking position, stacking height, tilt angle, and lateral misalignment of adjacent containers.

[0088] Collision avoidance safety calculation engine:

[0089] Anti-collision bowling logic: Based on the absolute position of the spreader obtained by GNSS and the misalignment information of adjacent containers obtained by scanning, the safe distance between the spreader / container to be placed and the adjacent stack is calculated in real time, the collision risk is predicted, and centimeter-level safety warnings are provided.

[0090] Deep pit protection logic: For deep stacking container grabbing operations, the dynamic distance between the lifting tool or the container to be grabbed and the edge of the stack side wall is calculated in real time. When the distance is less than the safety threshold, an early warning or braking is triggered immediately to prevent friction and collision.

[0091] (3) Precision alignment and operation assistance unit.

[0092] Container landing auxiliary calculation module: It compares the absolute coordinates of the spreader provided by high-precision GNSS with the precise spatial coordinates of the target container or stacking position lock hole obtained by 3D scanning, and calculates in real time the centimeter-level deviations of the spreader in terms of lateral, longitudinal and rotational yaw angles that need to be adjusted.

[0093] Guidance command generation module: Based on the deviation calculation results and operation logic, it automatically generates precise motion control commands or auxiliary guidance commands (such as PLC control quantities) to quickly guide the lifting device to align with the lock hole of the target box, thereby achieving box landing assistance.

[0094] (4) System control and human-computer interaction unit:

[0095] Data communication interface: Provides standard communication protocol interfaces (such as industrial Ethernet, Profibus, etc.) to send high-precision positioning data, safety warning signals and precise alignment instructions to the crane's PLC (programmable logic controller) or automation main control system to realize closed-loop motion control.

[0096] Operator display terminal: Provides a graphical interface in the driver's cab or remote control center, intuitively and dynamically displaying the real-time position of the spreader, the three-dimensional status of the stack, the safety zone, the collision warning level, and the container landing guidance information, assisting the operator in making efficient and safe operational decisions.

[0097] Furthermore, the steps for implementing integrated auxiliary container handling in this embodiment are as follows:

[0098] C1: Stacked 3D point cloud acquisition and preprocessing.

[0099] A high-performance lidar deployed on a crane trolley scans the container stacking operation area below at a fixed frequency in real time, capturing raw 3D point cloud data.

[0100] The point cloud preprocessing module filters, denoises, and segments the ground / foreground targets in the data. It then clusters the segmented foreground point clouds according to spatial proximity to identify key operational targets such as target containers, adjacent containers, and stack sidewalls.

[0101] C2: Accurate calculation of stacking status and safety parameters.

[0102] Container attitude extraction: For the target and adjacent container point clouds identified by clustering, the center coordinates, actual tilt angle, three-dimensional dimensions and lateral misalignment relative to the ideal stacking position are accurately calculated using fitting or feature extraction algorithms. A three-dimensional stacking yard map is established through positioning, and the container positions are synchronized between different yard cranes.

[0103] Safety clearance calculation (collision avoidance):

[0104] Anti-bowling calculation: Real-time calculation of the dynamic safety distance between the container to be placed (or the spreader) and the side wall of the adjacent container. When the distance is less than the preset safety threshold, a high-level warning signal is immediately triggered.

[0105] Deep pit protection calculation: When grabbing containers from deep stacking positions, the system calculates the closest dynamic distance between the spreader and the edge of the stack side wall in real time, and issues a warning or braking command when the distance is critical.

[0106] C3: Precision box placement assistance and instruction generation.

[0107] Alignment deviation calculation: The absolute spatial coordinates of the spreader obtained by GNSS in step 3 are compared in real time with the precise coordinates of the target container lock hole or ideal landing position obtained by 3D scanning in step 5, and the centimeter-level alignment deviations (ΔX, ΔY, θ) in horizontal and angular aspects are calculated.

[0108] Precise box-laying guidance commands: Based on the calculated deviation and combined with the kinematic model of the spreader, the optimal micro-motion control commands of the spreader (such as trolley movement, lifting / lowering, micro-rotation) are automatically generated to drive the crane PLC to perform precise adjustments and realize box-laying assistance.

[0109] C4: Automation control and data closed loop.

[0110] All high-precision positioning data, safety calculation results, and precise alignment commands are sent to the crane's main control PLC system in real time via a low-latency communication interface (such as industrial Ethernet).

[0111] After receiving the command, the PLC system executes high-precision motion control to realize safety logics such as anti-bowling and deep pit protection, and automatically completes the precise box placement operation.

[0112] The system continuously records key data such as positioning accuracy, number of safety warnings, and operation time for each container grabbing and placing operation, forming a closed-loop feedback loop for adaptive optimization of the algorithm and improvement of port operation performance.

[0113] Furthermore, the following section introduces the steps of integrated assisted container grabbing and placing in the yard from the perspective of algorithm flow.

[0114] D1: After calibrating the range of the trolley encoder or other positioning tools such as GNSS in the yard, activate this method when in the yard. Depending on the status of the trolley and spreader, enter different working modes:

[0115] When the trolley moves, it enters the "anti-bowling" mode and performs a "yard scan" at the same time.

[0116] When the trolley stops and the spreader twist lock opens, it enters the "automatic box grabbing" mode;

[0117] When the trolley stops and the spreader lock is closed, it enters the "automatic box placement" mode.

[0118] D2: To improve the accuracy of each step, clustering and noise removal will be performed. Statistical filtering will be used to remove noise by analyzing the distance distribution between each point and its neighbors, removing outliers whose distance distribution significantly deviates from the majority of points. An adaptive clustering parameter threshold-based clustering algorithm will be used. The input parameters of this algorithm include: the point cloud to be clustered P={p1,p2,...,pn}, distance coefficient α, base offset β, density coefficient γ, density search radius R_density, zero-prevention constant δ, minimum number of cluster points MIN_CLUSTER_SIZE (recommended value: 10), and maximum number of cluster points MAX_CLUSTER_SIZE (recommended value: 5000). The output of this clustering algorithm is: cluster set C={C1,C2,...,Ck}. The detailed steps of this clustering algorithm are as follows:

[0119] E1: System initialization.

[0120] 1.1 Construct the KD-Tree spatial index structure for point cloud P;

[0121] 1.2, Initialize the set of unvisited points U = P;

[0122] 1.3, Initialize an empty cluster list C = [];

[0123] 1.4 Initialize the point state flag array and mark all points as unvisited.

[0124] E2: Dynamic threshold calculation.

[0125] For each point pi in the point cloud P, perform the following calculations in parallel:

[0126] 2.1 Calculate the spatial distance feature: ri = sqrt(xi² + yi²), representing the horizontal distance to the radar coordinate origin;

[0127] 2.2 Calculate the base distance term: D_base(pi) = α × ri + β;

[0128] 2.3 Calculate local density features: With pi as the center, search for all nearest neighbor points within a spherical region with a radius of R_density, and count the number of nearest neighbor points N_neighbors(pi);

[0129] 2.4 Calculate the density correction term: ρ(pi) = γ / log(N_neighbors(pi) + δ);

[0130] 2.5 Calculate the final dynamic threshold: T_dynamic(pi) = D_base(pi) × ρ(pi).

[0131] E3: Adaptive Region Growing Clustering

[0132] 3.1 Main loop: When the unvisited point set U is not empty, repeat the following process;

[0133] 3.2 Seed point selection:

[0134] Randomly select a point p_seed from U as the initial seed point;

[0135] If the dynamic threshold T_dynamic(p_seed) of p_seed exceeds the reasonable range [0.1, 2.0], skip that point;

[0136] 3.3, New Cluster Initialization:

[0137] Create an empty cluster C_new and a point queue Q;

[0138] Add p_seed to queue Q and cluster C_new;

[0139] Remove p_seed from U and mark it as visited;

[0140] 3.4, Regional Growth Process:

[0141] When queue Q is not empty, execute:

[0142] Retrieve the current point p_current from the head of queue Q;

[0143] Get the dynamic threshold of p_current: T_current = T_dynamic(p_current);

[0144] Search for all unvisited adjacent points within a radius T_current around p_current;

[0145] For each adjacent node p_neighbor:

[0146] Calculate the semantic consistency score between two points (such as the angle between normal vectors, color similarity, etc.). If the semantic consistency is higher than the threshold, perform the following: add p_neighbor to queue Q, add p_neighbor to cluster C_new, remove p_neighbor from U, and mark it as visited.

[0147] 3.5, Clustering Validity Verification:

[0148] If the number of points in cluster C_new is within the range of [MIN_CLUSTER_SIZE, MAX_CLUSTER_SIZE], then calculate the geometric features of the cluster (bounding box size, point cloud density, etc.); if the geometric features meet the physical constraints, add C_new to the cluster list C; otherwise, re-mark the points in C_new as noise points.

[0149] E4: Post-processing optimization.

[0150] 4.1 Cluster merging: Merging small clusters that are spatially adjacent and semantically similar;

[0151] 4.2 Boundary Optimization: Optimizing cluster boundaries based on local surface continuity;

[0152] 4.3 Noise filtering: Remove abnormal clusters that do not clearly conform to the characteristics of the objects.

[0153] E5: Output of results.

[0154] Return the final cluster set C = {C1, C2, ..., Ck};

[0155] D3: When the spreader moves in the yard, it will perform anti-blow protection during horizontal movement. It will detect whether it is carrying a container and the direction of the spreader's movement to determine the alarm range to be detected. Different warnings will be given according to the distance of the obstacle from the spreader within the range. If the distance is less than 1m, a signal to limit the speed of the trolley to 50% will be sent to the PLC; if the distance is less than 0.1m, a signal to stop the trolley will be sent. The stop signal will last for 3 seconds. After 3 seconds, a signal to limit the speed of the trolley to 10% will be sent until the spreader is far away from the obstacle (50% speed limit if the distance is more than 0.1m but less than 1m, no speed limit if the distance is more than 1m).

[0156] D4: During spreader movement, activate the yard scanning function to record the yard information scanned by radar, interact with the positioning system to construct a 3D yard map, and synchronize it to different yard cranes. First, calibrate the origin of the 3D yard map and determine the corresponding positions of each radar and GNSS device. With yard scanning enabled, calculate the position of the GNSS antenna on the coordinate axis, inferred from its relative position to the origin. After calculating the antenna's position on the coordinate axis, determine the radar's position on the coordinate axis based on the relative position of the radar and the antenna. Finally, offset the point cloud of each frame according to the real-time position of the radar, ultimately converting all real-time point clouds into a single 3D yard point cloud map.

[0157] In 3D storage yard modeling, relying solely on positioning systems (such as GNSS, IMU, or encoders) often results in point cloud artifacts, misalignments, or gaps due to positioning error drift, signal obstruction, mechanical vibration, or encoder slippage. This solution introduces an ICP registration mechanism. The core logic of this technology shifts from "simply relying on positioning hardware" to "geometric feature-based correction," that is, iteratively solving for the optimal rotation matrix by finding corresponding point pairs within the overlapping area of ​​the frame to be registered and the reference map.

[0158] In practical applications, the algorithm further incorporates Generalized ICP (G-ICP), utilizing the local geometric distribution (covariance matrix) of the yard ground, retaining walls, and material surface to construct "surface-to-surface" constraints. By increasing the error weights perpendicular to the geometric surface, it forces the point cloud to achieve precise adsorption on the physical structure. This fusion of "initial prediction by the positioning system + fine compensation by geometric features" effectively eliminates the cumulative errors caused by sensor drift, ensuring the consistency and computational accuracy of the yard's 3D digital twin model at the global scale.

[0159] The specific steps are as follows:

[0160] F1: Spatial synchronization and overlapping area extraction.

[0161] Using initial extrinsic values ​​provided by the positioning system (GNSS, IMU, or encoder), the current scan frame is roughly projected onto the global coordinate system. Based on the overlapping bounding boxes (AABB) of the two frames' poses, the geometrically overlapping regions are automatically identified and extracted, and invalid data in the far-end non-overlapping parts is discarded. This step significantly reduces the search space using positioning data, providing a highly reliable "initial pose" for precise registration.

[0162] F2: Ambient noise filtering and adaptive downsampling.

[0163] To address the high density of dust, water mist, and moving robotic arms (such as bucket wheel excavators and belt conveyors) in stockyard operations, a statistical outlier filter (SOR) is used to identify and remove spatial discrete noise. Subsequently, the point cloud is downsampled using an adaptive voxel grid algorithm. While ensuring that key geometric features such as stockpile edges and retaining wall edges are not lost, the point cloud density is adjusted to a computationally balanced range to improve the algorithm's real-time performance.

[0164] F3: Anisotropic covariance modeling.

[0165] By performing K-neighborhood feature analysis on the point cloud of the overlapping area, this scheme further introduces a semantic feature classification operator. Principal component analysis (PCA) is used to obtain feature values ​​within the neighborhood of the sampling points. A local geometric feature descriptor is constructed (λ1≥λ2≥λ3≥0, representing the first, second, and third principal component feature values ​​respectively; λ1 is the extension in the longest direction, λ2 is the extension in the second longest direction, and λ3 is the extension in the thickness direction):

[0166] Linearity:

[0167]

[0168] Surface flatness (Planarity):

[0169]

[0170] Scattering / Singularity:

[0171]

[0172] Based on the values ​​of these three parameters, the point cloud is classified into: planar point sets (container side / top), edge point sets (container edges), and key singularity sets (container corner lock holes). This step provides the semantic basis for subsequent differential weighted registration. The classification rules are as follows:

[0173] First, the system uses scattering intensity to identify key singularity sets. Since the point cloud distribution in container lockholes and corner fitting areas is typically chaotic and lacks clear planar or linear features, the system uses scattering intensity as the core criterion. When the scattering intensity of a sampling point exceeds a preset sensitivity threshold (e.g., 0.2), and the scattering intensity of that point is numerically greater than both its linear elasticity and surface flatness, the system prioritizes identifying that point as a key singularity. This type of point set mainly corresponds to the corner fittings and internal lockhole structures of the container, serving as a crucial reference for achieving high-precision container loading.

[0174] After excluding the aforementioned singularity features, the system further filters the linear geometric attributes of the remaining point cloud. If a point is not identified as a singularity and its linear elasticity value exceeds a preset linear feature threshold (e.g., 0.5), the system classifies it as an edge point set. This type of point set precisely corresponds to the twelve external contour edges of the container, providing crucial boundary constraints for the lateral and longitudinal displacements of the crane during container handling.

[0175] For the remaining point cloud that does not belong to the singularity set or edge point set, the system will make a final classification based on the dominance of its planar attributes. If the surface flatness value of a point is dominant among the three geometric descriptors and exceeds a preset planar feature threshold (e.g., 0.5), it is determined to be a planar point set. This type of point set corresponds to the large-area box panels, roof panels, and yard floors of containers. In subsequent calculations, the system will adopt a weight suppression strategy for this point set to avoid large-area planar features from excessively dominating the algorithm's convergence direction, thereby ensuring the accuracy of edge and keyhole feature localization.

[0176] Construct the anisotropic covariance matrix and perform the following transformation:

[0177]

[0178] V is a rotation basis composed of eigenvectors, and ϵ is a minimal variance set along the normal direction. This step logically fits each sampling point as a tiny "geometric disk," giving the algorithm the degree of freedom to slide in the direction parallel to the material surface, while strictly limiting the offset in the vertical direction.

[0179] F4: Establish a multidimensional constraint cost function.

[0180] The objective optimization function is constructed based on Mahalanobis distance, using the prior pose provided by the localization system as a soft constraint term, and together with the geometric residual term, to construct the cost function:

[0181]

[0182] in, The cost function represents the total energy cost after registering the point clouds of two frames. The algorithm iteratively optimizes the transformation matrix T (which includes rotation R and translation t) to minimize this value, thereby achieving perfect geometric alignment between the current frame and the reference map.

[0183] in, It is a residual vector: representing points in the source point cloud. After pose transformation T and projection, the corresponding point in the target point cloud The spatial displacement vector between the two point clouds. It intuitively reflects the size of the physical "gap" between the two point clouds at the current pose.

[0184] in, It is the Mahalanobis distance weight operator: and These are the local covariance matrices of the target point and the source point, respectively (representing the local geometry of the surface). This term transforms the simple Euclidean distance into Mahalanobis distance. By applying directional weighting to the residual vector, a very high error penalty weight is assigned to the direction perpendicular to the material surface (normal direction), while the displacement constraint parallel to the material surface is relaxed. This design enables the algorithm to logically achieve precise "surface-to-surface" adsorption, effectively correcting the drift in the height and horizontal directions caused by mechanical vibration or slippage in the positioning system.

[0185] in, (This is a robust kernel function): Huber Loss or Cauchy kernel functions are typically used. Its core logic is to dynamically adjust the weights of outlier matching points: when the residual of a certain point becomes too large due to dust noise, equipment obstruction, or dynamic material collapse, ρ will limit its contribution to the total cost, preventing local outliers from "distorting" the overall pose model.

[0186] in, Achieving global summation: This involves globally summing the errors of all valid corresponding point pairs within the overlapping region. Through collaborative optimization of massive number of point pairs, it ensures that the final transformation matrix T is the optimal solution under global geometric consistency constraints, thereby eliminating faults and ghosting.

[0187] in, It's weight. The value selection logic is as follows: when the sampling point is identified as a key singularity (keyhole region) [5,10], the algorithm is forced to prioritize aligning the keyhole features; when the sampling point is identified as an edge point [2,5], and when the sampling point is a large-area planar point [0.5,1]. By significantly enhancing the "voice" of keyhole and edge features in registration, the problem of traditional G-ICP being masked by the excessive proportion of planar points is solved, improving the registration accuracy from decimeter level to centimeter level and ensuring physical alignment accuracy.

[0188] F5: Iterative optimization and convergence determination.

[0189] The Levenberg-Marquardt (LM) nonlinear optimization algorithm is used to solve for the transformation matrix. During the iteration process, the KD-Tree is used to dynamically update the point pair associations, continuously compressing the Mahalanobis distance. A kinematic boundary constraint factor based on the physical characteristics of the crane is introduced. When solving for the transformation matrix T (containing the translation vector t=[dx,dy,dz]), the following hard constraints are added:

[0190]

[0191] in, and These represent the current maximum physical speeds of the crane trolley / crane, with x and y representing the directions. and These represent the maximum acceleration of the current crane trolley / crane, with x and y representing the directions. This is the sampling time step. This constraint, acting as a 'physical verification layer,' can eliminate non-physical displacement solutions (i.e., so-called 'jumps') caused by sensor flicker or strong environmental interference, ensuring that the output control commands are within the acceptable range of the crane's mechanical actuators, thus enhancing the inherent safety of the system.

[0192] F6: Pose Correction and Consistency Verification.

[0193] The final optimized transformation matrix is ​​output and applied to the current frame point cloud, while the coordinates recorded by the original positioning system are backtracked and corrected. The alignment accuracy is verified by calculating the root mean square error after registration. If the error is within the acceptable range, the corrected point cloud is merged into the local submap to alleviate ghosting and discontinuities caused by encoder slippage or vibration, thus completing the dynamic update of the high-precision stockpile map.

[0194] D5: When the spreader is in the yard and descending, anti-collision bowling and auxiliary container grabbing / release are activated to prevent collisions between the spreader and the transported container. The system also scans the yard information directly below in real time, analyzes the position of the container directly below, and assists the spreader in correcting container grabbing / release. In "automatic grabbing" and "automatic releasing" modes, the system monitors the distance below the spreader or the container being grabbed, as well as the distance to obstacles. If the distance is less than 0.5m, a 10% speed limit signal is sent to the PLC. During assisted container grabbing / release, the algorithm performs clustering and noise removal, and locates the position of each cluster. It selects the container point cloud directly below and around the container for container positioning. For container position and corner positioning, based on the upper surface of the clustered container, it obtains the four maximum and minimum values ​​of all points to determine a horizontal rectangle. The center of this rectangle is obtained, and then the corner positions are determined based on the container's model and specifications.

[0195] In detail, this involves obtaining the four horizontal values ​​X_max, X_min, Y_max, and Y_min of the upper surface, obtaining the center of the upper surface of the container at (X_max + X_min) / 2 and (Y_max + Y_min) / 2, determining the container model, and then locating the four corners according to the standard model size. For example, if the container width is 2.438m, assuming (Y_max + Y_min) / 2 is the center of the width, the positions of the four corners in the width direction are approximately 1.219m to the left and right of (Y_max + Y_min) / 2, and the same applies to the length direction.

[0196] D6: When grabbing and placing containers, the system provides additional functions for deep pit operation and container landing assistance. It determines whether the operation is to be carried out in a deep pit by scanning the information of the yard below. If it enters a deep pit, the speed of the spreader can be limited. When grabbing and placing a container is about to approach the container directly below for landing operation, it will slow down to prevent the container from colliding with it due to excessive speed.

[0197] If the lifting equipment enters a deep pit for operation, a signal will be issued to limit the speed of the lifting equipment and trolley by 10%.

[0198] If the spreader does not enter the pit, it will send the forward, backward, left and right deviations and angular deviations of the spreader relative to the container or empty space below to the PLC in real time, so as to adjust the spreader's position and posture to achieve automatic container placement.

[0199] In addition to the above signals, the system will also send the distance between the lower edge of the spreader and the container or ground directly below to the PLC.

[0200] Example 2

[0201] This embodiment describes the on-site deployment method of the integrated auxiliary container grabbing and placing method in the yard described in Embodiment 1.

[0202] 1. Base station module: The GNSS antenna is deployed on the roof of a nearby unobstructed high-rise building or the top of a signal tower. It is usually installed on the roof of the port remote control center building. The base station receiver and controller are usually installed in the electrical room of the remote control center building and connected to the port network via an Ethernet switch.

[0203] 2. Mobile station module: GNSS antennas are deployed at the four corners of the crane's main frame and trolley frame. The mobile station receiver and controller are installed in the electrical cabinet of the trolley frame and connected to the IPC in the main frame's electrical room via network cables. The IPC is then connected to the port network via an Ethernet switch.

[0204] 3. High-performance 3D scanning equipment (LiDAR / visual scanner): Strategically installed below the crane trolley (such as on top of the spreader or at the bottom of the trolley frame), selecting the best observation point to ensure a comprehensive, blind-spot-free real-time scanning capability of the container stacking area below.

[0205] 4. Operator display terminal: A high-resolution display is installed in front of the crane operator's cab or the control panel of the remote control center, providing an intuitive graphical user interface and displaying key operation information in real time.

[0206] 5. Data Processing and Control System: An industrial-grade server is installed in the electrical room as the core of the data processing system. This server is equipped with a high-performance CPU and GPU to run complex point cloud processing and recognition algorithms. The server connects to the PLC control system of the rail-mounted gantry crane via standard industrial communication protocols (such as Modbus TCP) and has the ability to send commands to the automation system.

[0207] 6. System Network and Power Supply: All equipment is connected through the industrial network inside the rail-mounted gantry, forming an independent and stable local area network. The system's power is supplied by its own power supply system, ensuring synchronization with the port's operational rhythm and achieving 24 / 7 uninterrupted operation.

[0208] Example 3

[0209] The integrated auxiliary container handling system for stockyards described in this invention includes:

[0210] The yard 3D map creation and update unit is used to create a yard 3D map and acquire 3D point cloud data of the container stacking operation area in real time during the movement of the spreader, which is used to update the yard 3D map.

[0211] The auxiliary container grabbing and placing unit is used to perform anti-bowling and deep pit protection calculations based on the three-dimensional map of the yard during the container grabbing and placing process, calculate the dynamic safety distance around the container and / or lifting gear, and issue an early warning signal when the dynamic safety distance is less than a threshold.

Claims

1. A comprehensive auxiliary container handling method for stockyards, characterized in that, Includes the following steps: A 3D map of the yard is established, and 3D point cloud data of the container stacking operation area is acquired in real time during the movement of the spreader to update the 3D map of the yard. During the container handling process, anti-bowling and deep pit protection calculations are performed based on the three-dimensional map of the yard. The dynamic safety distance around the container and / or spreader is calculated, and an early warning signal is issued when the dynamic safety distance is less than the threshold.

2. The integrated auxiliary container handling method for stockyards according to claim 1, characterized in that, A 3D map of the storage yard was created using GNSS positioning. A lidar is installed on the crane trolley to acquire three-dimensional point cloud data of the container stacking operation area. The position of the lidar in the three-dimensional map of the yard is determined according to the relative position of the lidar and the GNSS antenna. Each frame of three-dimensional point cloud data acquired by the lidar is offset to complete the fusion of the three-dimensional point cloud data and the three-dimensional map of the yard, resulting in a dynamically updated three-dimensional map of the yard.

3. The integrated auxiliary container handling method for stockyards according to claim 2, characterized in that, The process of offsetting each frame of 3D point cloud data acquired by the lidar to fuse the 3D point cloud data with the 3D map of the storage yard includes: 3D point cloud data offset is performed using the ICP registration mechanism: Using GNSS positioning data as the initial value of the external parameter, the current frame's 3D point cloud data is projected onto the global coordinate system of the 3D map of the storage yard to identify geometrically overlapping areas; The 3D point cloud data is downsampled using an adaptive voxel mesh algorithm to adjust the point cloud density to a computational equilibrium range, thus obtaining the first point cloud data. Construct a covariance matrix based on the first point cloud data of the overlapping region of the set; GNSS positioning data is used as a soft constraint term, which, together with geometric residual terms, is used to construct a cost function. The transformation matrix is ​​iteratively optimized to minimize the value of the cost function. The optimized transformation matrix is ​​applied to the current frame's 3D point cloud data to complete the 3D point cloud data offset. By fusing each frame of offset 3D point cloud data with the 3D map of the storage yard, a dynamically updated 3D map of the storage yard is obtained.

4. The integrated auxiliary container handling method for stockyards according to claim 3, characterized in that, The cost function is: ； in, It is the i-th sampling point in the first point cloud data. It is the i-th sampling point in the first point cloud data after offset. The transformation matrix; It is the Mahalanobis distance weight operator. and These are the covariance matrices of the first point cloud data and the offset first point cloud data, respectively. It is a robust kernel function; As weight; The logic for determining the value is as follows: Semantic feature classification is performed on the first point cloud data of the overlapping region of the set, including planar point set, edge point set, and key singularity set; when When identified as a key singularity ,when When identified as an edge point ,when When it is a large area planar point .

5. The integrated auxiliary container handling method for stockyards according to claim 3, characterized in that, Principal component analysis is used to obtain the eigenvalues ​​of the first point cloud data in the neighborhood, and the linear elasticity, surface flatness and scattering are calculated. The first point cloud data is divided into planar point sets, edge point sets, and key singularity sets based on the linear elasticity, surface flatness, and scattering. If the scattering index of the point cloud data exceeds the first threshold, it is determined to be a key singularity; otherwise, the judgment is made based on the linear elasticity index of the point cloud data. If the linear elasticity index of the point cloud data exceeds the second threshold, it is determined to be an edge point; otherwise, the judgment is made based on the surface flatness index of the point cloud data. If the surface flatness index of the point cloud data exceeds the third threshold, it is determined to be a planar point.

6. The integrated auxiliary container handling method for stockyards according to claim 1, characterized in that, GNSS positioning data is acquired by deploying GNSS positioning units in the yard, which is then used to create a three-dimensional map of the yard. The GNSS positioning unit includes a base station module and a rover module; The base station module includes a first GNSS antenna, a first network switching device, a base station receiver, and a base station controller, which are used to capture multi-constellation satellite signals and forward differential data sources and reference references to the rover module; The rover module includes a second GNSS antenna, a second network switching device, an industrial control computer, a rover receiver, and a rover controller, which are used to perform RTK calculations.

7. A comprehensive auxiliary container handling system for a storage yard, characterized in that, include: The yard 3D map creation and update unit is used to create a yard 3D map and acquire 3D point cloud data of the container stacking operation area in real time during the movement of the spreader, which is used to update the yard 3D map. The auxiliary container grabbing and placing unit is used to perform anti-bowling and deep pit protection calculations based on the three-dimensional map of the yard during the container grabbing and placing process, calculate the dynamic safety distance around the container and / or lifting gear, and issue an early warning signal when the dynamic safety distance is less than a threshold.

8. The integrated auxiliary container handling system for stockyards according to claim 7, characterized in that, In the yard 3D map creation and update unit, a yard 3D map is created through GNSS positioning; a lidar is set on the crane trolley to acquire 3D point cloud data of the container stacking operation area; the position of the lidar in the yard 3D map is determined according to the relative position of the lidar and the GNSS antenna; each frame of 3D point cloud data acquired by the lidar is offset to complete the fusion of the 3D point cloud data and the yard 3D map, and a dynamically updated yard 3D map is obtained. In the auxiliary grabbing and placing box unit, the ICP registration mechanism is used for 3D point cloud data offset: Using GNSS positioning data as the initial value of the external parameter, the current frame's 3D point cloud data is projected onto the global coordinate system of the 3D map of the storage yard to identify geometrically overlapping areas; The 3D point cloud data is downsampled using an adaptive voxel mesh algorithm to adjust the point cloud density to a computational equilibrium range, thus obtaining the first point cloud data. Construct a covariance matrix based on the first point cloud data of the overlapping region of the set; GNSS positioning data is used as a soft constraint term, which, together with geometric residual terms, is used to construct a cost function. The transformation matrix is ​​iteratively optimized to minimize the value of the cost function. The optimized transformation matrix is ​​applied to the current frame's 3D point cloud data to complete the 3D point cloud data offset. By fusing each frame of offset 3D point cloud data with the 3D map of the storage yard, a dynamically updated 3D map of the storage yard is obtained.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the integrated auxiliary container grabbing and placing method for the yard according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the integrated auxiliary container grabbing and placing method for the yard according to any one of claims 1-6.

Images